Nowadays,multi-label classification methods are increasingly required by modern applications,such as protein function classification, music categorization and semantic scene classification. This paper introduces the task of multi-lablel classification, organizes the sparse related literature into a structured presentation ann performs comparative experimental results of certains multi-label classification methods. It also contributes the definition of concepts for the quantification of the multi-label nature of a data set.
This paper aims to serve as a starting point and reference for researchers interested in multi-label classification. The main contributions are: a) a structured presentation of the sparse literature on multi-label classification methods with comments on their relative strengths and weaknesses and when possible the abstraction of specific methods to more general and thus more useful schemata, b) the introduction of an undocumented multi-label method, c) the definition of a concept for the quantification of the multi-label nature of a data set, d) preliminary comparative experimental results about the performance of certain multi-label methods.
这段话介绍写这篇文章的目的，总共有四个：
系统介绍当前已有方法的原理及优缺点；
介绍一些还没有文档记录的方法的性能；
介绍多类标分类问题的度量准则；
比较现有方法的优缺点，以及实验结果。
问题的相关工作
A task that also belongs to the general family of supervised learning and is very relevant to multi-label classification is that of ranking. In ranking the task is to order a set of labels L, so that the topmost labels are more related with the new instance. There exist a number of multi-label classification methods that learn a ranking function from multi-label data. However, a ranking of labels requires post-processing in order to give a set of labels, which is the proper output of a multi-label classifier.
Multi-Label Classification: An Overview
----------Grigorios Tsoumakas----------Ioannis Katakis
Nowadays,multi-label classification methods are increasingly required by modern applications,such as protein function classification, music categorization and semantic scene classification. This paper introduces the task of multi-lablel classification, organizes the sparse related literature into a structured presentation ann performs comparative experimental results of certains multi-label classification methods. It also contributes the definition of concepts for the quantification of the multi-label nature of a data set.Abstract:
综述大概都是这样写的，感觉由Grigorios Tsoumakas和他的老师Ioannis Katakis来写这个综述还是很有说服力的，因为他们基于现有多类标分类方法做的mulan系统非常有用，而且涵盖面很广。解读：
前面写的是问题的起源，看过我前面写的您就能知道，这不是本文的重点，本文是组织目前典型方法的相关文档，有组织有分类的进行介绍，最后演示这些方法的结果比较。
除此之外，还介绍一些相关的概念，还有度量标准，比如基于类标的，基于实例，基于二元的，基于多元的结果度量方法，对我们多类标问题的学习和研究有很重要的意义。
文章解读：
文章简介
多类标问题与传统分类问题的区别：主要是类标号的数量不同，多类标问题各类标号之间有相互依赖，文章中举了一个电影《达芬奇密码》的例子，很多文章都有引用。This paper aims to serve as a starting point and reference for researchers interested in multi-label classification. The main contributions are: a) a structured presentation of the sparse literature on multi-label classification methods with comments on their relative strengths and weaknesses and when possible the abstraction of specific methods to more general and thus more useful schemata, b) the introduction of an undocumented multi-label method, c) the definition of a concept for the quantification of the multi-label nature of a data set, d) preliminary comparative experimental results about the performance of certain multi-label methods.
这段话介绍写这篇文章的目的，总共有四个：
问题的相关工作
A task that also belongs to the general family of supervised learning and is very relevant to multi-label classification is that of ranking. In ranking the task is to order a set of labels L, so that the topmost labels are more related with the new instance. There exist a number of multi-label classification methods that learn a ranking function from multi-label data. However, a ranking of labels requires post-processing in order to give a set of labels, which is the proper output of a multi-label classifier.我们着重看一个问题就是和传统的有监督学习方法练习紧密的方法就是多类标分类中的ranking方法，它的主要思想是给类标集合排序，排名越靠前的类标取值和样例的相关性越大。多类标分类问题的ranking方法有很多种，然而这些方法都需要对预测数据集
实验分析